A study on analytical transform kernel derivation for Versatile Video Coding (VVC)
- Author(s)
- 산딥 쉬레스따
- Issued Date
- 2020
- Abstract
- Versatile Video Coding (VVC)의 표준화에서는 DCT-2 (Discrete Cosine Transform-2), DST-7 (Discrete Sine Transform-7), DCT-8 (Discrete Cosine Transform-8)이 주변환커널로 사용되고 있다. 그러나 이들 변환 커널과 함께 DST-4 (Discrete Sine Transform-4)와 DCT-4 (Discrete Cosine Transform-4)도 역시 그 압축 성능과 구현성 복잡도 이득으로 인해 표준화 과정 중 주 변환 커널로 고려되어왔다. 구현 복잡도의 이점으로 인해 DST-4와 DCT-4는 DST-7과 DCT-8 변환 커널의 대안으로 사용될 수 있다. 그러나 이러한 다양한 종류의 변환 커널과 크기는를 정수 계수로 저장하기 위해선느 많은 메모리가 필요하며 이 것은 VVC 비디오 코덱 중요한 복잡도 문제가 될 수 있다. 이러한 문제를 해결하기 위하여 본 학위 논문에서는 unified sparse 행렬의 개념을 제안하고, 이 행렬을 이용하여 서로 다른 크기의 변환 커널 매트릭스(DCT-2, DST-7, DCT-8, DST-4, DCT-4)를 간단한 수학적 연산을 통해 얻을 수 있다. 제안하는 unified sparse matrix는 unified-DST-3 행렬과 grouped-DST-7 행렬의 두 부분으로 구성된다. unified-DST-3 행렬은 서로 다른 크기의 변환 커널 DCT-2, DST-4 및 DCT-4 변환 커널을 유도 사용되며, gruped-DST-7 행렬은 다양한 크기의 DST-7 및 DCT-8 변환 커널을 도출하는 데 사용된다. grouped-DST-7 행렬에서 32 픽셀 크기의 DST-7에 대해 두 가지 접근방식(Approach 1과 Approach 2)을 이용할 수 있다. 실험은 공통 테스트 조건(CTC)에 따라 VTM-3.0 참조 소프트웨어에서 수행되었고 제안하는 방법의 BDBR 결과는 AI와 RA 조건에 대해 압축 성능에 영향을 주지 않았고으나 기존의 방법에 비해 79.85%의 메모리 저감 효과를 얻을 수 있었다. |In the standardization of Versatile Video Codec (VVC), DCT-2 (Discrete Cosine Transform-2), DST-7 (Discrete Sine Transform-7), and DCT-8 (Discrete Cosine Transform-8) are being regarded as the major transform kernels. However, along with the defined transform kernels, DST-4 (Discrete Sine Transform-4) and DCT-4(Discrete Cosine Transform-4) are also counted as the basic transform kernels. Usually, DST-4 and DCT-4 can be used as the replacement of the DST-7 and DCT-8 transform kernels respectively for their effectiveness in smaller resolution sequences. While storing all those transform kernels, memory usage is regarded as the major issue. To deal with this scenario, a common sparse unified matrix concept is introduced in this thesis paper from which any point transform kernel matrix can be obtained i.e. DCT-2, DST-7, DCT-8, DST-4 and DCT-4 after some mathematical operations. The defined common sparse unified matrix is composed up of two parts: unified DST-3 matrix and grouped DST-7 matrix. Unified DST-3 matrix is used to derive different block size DCT-2, DST-4 and DCT-4 transform kernels whereas the grouped DST-7 matrix is used to derive different block size DST-7 and DCT-8 transform kernels. In the grouped DST-7 matrix concept, two approaches (approach 1 and approach 2) has been introduced for the 32-point DST-7 to make 32-point DST-7 transform kernel compatible with the proposed algorithm. Similarly, the tuning of one element of the certain row of the DST-7 transform kernel affects other rows of DST-7 transform kernel is also shown in the proposed grouping concept of the DST-7 transform kernel matrix. The transform kernels used for simulation are DCT-2, DST-7 and DCT-8. The simulation is conducted in VTM-3.0 reference software under the common test condition (CTC). The BDBR results of the proposed methods showed no significant loss for AI and RA. However, for LDB, approach 1 showed a gain of 0.21% for V-chroma and less significant loss of 0.09% for U-chroma whereas approach 2 showed a gain of 0.05% for V-chroma and loss of 0.30% for U-chroma. Similarly, the proposed method has no significant impact on the encoding and decoding time which shows that there is no complexity issue in the proposed method even with 79.85% memory reduction.
- Alternative Title
- Versatile Video Coding(VVC) 비디오 코딩을 위한변환 커널 유도 방법에 관한 연구
- Alternative Author(s)
- Sandeep Shrestha
- Department
- 일반대학원 정보통신공학과
- Advisor
- Bumshik Lee
- Awarded Date
- 2020-02
- Table Of Contents
- List of figures iv
List of tables vi
한글요약 ix
Abstract xi
1. Introduction 1
1.1 Background 1
1.2 Objectives 6
1.3 Motivation 7
1.4 Thesis Layout 9
2. Related Works 10
2.1 Compound Orthonormal Transform (COT) 10
2.2 Unified Matrix for Transform 11
2.3 Adjustment Stages 12
2.4 Transform Adjustment Filter (TAF) 14
3. Proposed Method 16
3.1 Common Sparse Unified Matrix 16
3.2 Unified DST-3 Matrix (U) 17
3.2.1 DST-3 and DCT-2 Relationship 18
3.2.2 The Proposed Derivation Method of the DCT-2 Kernel 19
3.3 The Proposed Grouped DST-7 Transform Kernel 23
3.3.1 Grouping Concept 24
3.3.2 Specific Rows of Different Point DST-7 27
3.3.3 Derivation of DCT-8 29
3.4 Permutation Matrix 29
3.4.1 DST-7 Transform Kernel Pattern 30
3.4.2 Algorithm for Generation of Permutation Matrix (G) 32
3.4.3 First-half Algorithm 34
3.4.4 Second-half Algorithm 36
3.4.5 Alternate Method 36
3.5 Exceptional Cases 39
3.5.1 Exceptional Condition in Proposed Grouping Concept 39
3.5.2 Approach 1 40
3.5.3 Approach 2 41
3.6 DST-7 Transform Kernel Fast Algorithm 42
3.6.1 DST-7/DCT-8 Fast Algorithm 43
3.6.2 Fast Algorithm Implementation on The Proposed DST-7 47
4. Simulation Results and Discussions 50
4.1 Simulation Results of Proposed methods 50
4.2 Approach 1 and Approach 2 Comparison 63
4.3 Comparison Between Proposed Method and COT 63
4.4 Comparison Between Proposed Method and Unified Matrix 65
4.5 Proposed Method and Adjustment Stages Comparison 66
4.6 Proposed Method and TAF Comparison 67
5. Conclusion 70
Acknowledgment 71
References 72
- Degree
- Master
- Publisher
- Chosun University
- Citation
- 산딥 쉬레스따. (2020). A study on analytical transform kernel derivation for Versatile Video Coding (VVC).
- Type
- Dissertation
- URI
- https://oak.chosun.ac.kr/handle/2020.oak/14033
http://chosun.dcollection.net/common/orgView/200000279047
-
Appears in Collections:
- General Graduate School > 3. Theses(Master)
- Authorize & License
-
- AuthorizeOpen
- Embargo2020-02-26
- Files in This Item:
-
Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.